Lower limb assistive technologies can help transfemoral amputees restore locomotion and perform diverse daily activities. The ENcyclopedia of Able-bodied Bilateral Lower Limb Locomotor Signals (ENABL3S) serves as a benchmark dataset, capturing neuro-mechanical signals via wearable sensors during locomotor tasks such as sitting, standing, walking and ascending or descending stairs and ramps, where these tasks constitute an imbalanced target variable. This paper aims to offer a unified perspective on both predictive accuracy and interpretability by applying two complementary methodologies to the same complex dataset. Real time locomotor prediction is accomplished using Recurrent Neural Networks (RNN), including Long Short-Term Memory Networks (LSTM) and Gated Recurrent Units (GRU). In parallel, the Latent Budget Tree, a probabilistic machine learning model, is employed to optimize feature settings for all tasks, including underrepresented classes, providing valuable insights into locomotion dynamics. This dual perspective highlights the need for personalized exoskeletons and effective management of complex motor transitions.